Spectral Feature Mapper
Spectral Feature Mapper identifies a spectral peak or feature from a user-selected image region and maps that feature throughout the hyperspectral image.
Overview
Spectral Feature Mapper identifies a spectral peak or feature from a user-selected image region and maps that feature throughout the hyperspectral image.
ROI -> Mean Spectrum -> Feature Selection -> Feature Maps The tool can generate three spatial maps:
Click Select ROI and draw a rectangle over a representative region of the hyperspectral image.
IDCubePro calculates the mean spectrum of all pixels contained inside the selected region.
Choose an ROI that represents the material, tissue, sample region, or spectral population whose feature you want to characterize.
The ROI can be repositioned or resized. The mean spectrum updates after the ROI is moved.
The upper-right plot displays the average spectrum calculated from the selected ROI.
The original spectrum and, when enabled, the smoothed spectrum are displayed together.
Using the mean ROI spectrum reduces pixel-level noise and makes spectral features easier to identify.
Click Auto Peak after selecting an ROI.
IDCubePro identifies the strongest spectral maximum in the current mean spectrum.
The approximate feature width is estimated from the half-height crossing points around that maximum.
The resulting feature range can then be adjusted manually.
Auto Peak identifies the strongest peak according to the current ROI spectrum.
It does not determine whether that peak is the scientifically relevant spectral feature.
Always inspect the spectrum and adjust the range when necessary.
Feature Start and End define the wavelength interval used for feature analysis.
You can enter these values directly.
The values are snapped to the nearest wavelengths available in the hyperspectral dataset.
The Center slider changes the approximate center position of the selected feature interval.
The value is automatically matched to the nearest available wavelength.
The Half-width slider controls how far the selected feature interval extends on each side of its center.
A wider range includes more spectral bands.
A narrower range focuses on a smaller feature.
The selected range is displayed directly on the spectrum.
Vertical boundary lines indicate the beginning and end of the selected interval.
The shaded region represents the signal used for feature integration.
Enable Smooth spectrum to apply Savitzky-Golay smoothing.
Smoothing can reduce high-frequency noise while preserving broad peak structure.
Available smoothing windows are 5, 7, 9, and 11 spectral samples.
Larger windows produce stronger smoothing.
Excessive smoothing can alter narrow spectral features, so use a window appropriate for the spectral sampling and expected feature width.
Local Background Subtraction
Enable Subtract local linear background to estimate the signal above a local baseline.
The baseline is calculated as a straight line connecting the signal at the beginning and end of the selected feature range.
For every wavelength inside the selected range:
Corrected signal = measured signal - local background Negative corrected values are set to zero.
This correction can provide a better estimate of peak height and integrated peak area when a feature sits on a sloping spectral background.
When To Disable Local Background
Disable background subtraction when the absolute signal within the selected spectral interval is the desired measurement.
Whether background subtraction is scientifically appropriate depends on the type of spectral data and the interpretation of the feature.
Four metrics are displayed after a feature range has been selected.
Maximum background-corrected signal within the selected wavelength range.
Integrated spectral signal within the selected range.
When local background correction is enabled, the area represents the integrated signal above the estimated local baseline.
Wavelength at which the maximum corrected signal occurs.
Difference between the selected end and start wavelengths.
Click Make Maps after selecting the desired feature range.
IDCubePro analyzes the same spectral interval for every image pixel.
If smoothing is enabled, the spectral feature cube is smoothed along the wavelength dimension before map calculation.
If local background subtraction is enabled, a pixel-specific local baseline is subtracted before calculating the maps.
The Peak Intensity map shows the maximum signal within the selected feature interval for every image pixel.
When local background correction is enabled, this is the maximum signal above the estimated baseline.
This map can identify regions with strong expression of a particular spectral feature.
The Peak Area map integrates the signal across the selected wavelength range for every image pixel.
Area can be more robust than a single peak intensity when a spectral feature is broad or when its maximum position varies slightly across the image.
The Peak Wavelength map shows the wavelength of maximum signal within the selected interval for every image pixel.
This can be useful for identifying spatial shifts in spectral peak position.
Peak-position maps should be interpreted cautiously in low-signal regions because noise can affect the position of the local maximum.
Peak intensity measures the maximum value at one wavelength.
Peak area incorporates information across the complete selected feature interval.
Peak area is often useful for broad spectral features, while peak intensity can be useful for narrow or well-defined spectral maxima.
Peak Wavelength Interpretation
Changes in peak wavelength may indicate changes in chemical environment, composition, scattering, binding state, material properties, or other spectral effects depending on the experiment.
However, apparent peak shifts can also result from noise, low signal, spectral sampling, preprocessing, or overlapping features.
Click Save Maps to save the generated maps as a MATLAB MAT file.
- Whether local background correction was used
- Whether spectral smoothing was used
Feature mapping depends on the spectral quality of the hyperspectral cube.
Appropriate preprocessing may include:
- Reflectance or reference correction
- Removal of noisy wavelengths
- Normalization when scientifically appropriate
Use preprocessing that is appropriate for the imaging modality and scientific objective.
The ROI should contain enough representative pixels to provide a stable mean spectrum.
Very small ROIs may be strongly affected by pixel noise.
Very heterogeneous ROIs may average several different spectral populations and obscure the feature of interest.
1. Load and preprocess the hyperspectral dataset.
2. Open Spectral Feature Mapper.
3. Select a representative ROI.
4. Inspect the mean ROI spectrum.
5. Enable smoothing if useful.
7. Inspect the automatically selected feature.
8. Adjust Start, End, Center, or Half-width if needed.
9. Decide whether local background subtraction is appropriate.
10. Inspect Height, Area, Peak, and Width.
12. Compare Peak Intensity, Peak Area, and Peak Wavelength maps.
13. Save the resulting maps.
Auto Peak Selects The Wrong Feature
The strongest peak may not be the feature of interest. Enter the desired feature range manually or move the Center slider.
Use a larger representative ROI, enable smoothing, or preprocess noisy spectral bands before analysis.
The Feature Area Looks Too Small
The selected range may be too narrow, or local background subtraction may remove a substantial baseline contribution.
The Feature Area Looks Too Large
The selected range may include neighboring spectral features or excessive background.
Peak Wavelength Map Looks Noisy
Peak position becomes unstable when signal is weak. Consider masking low-signal pixels or using the intensity/area maps to identify reliable regions.
Maps Are Dominated By Background
Consider masking irrelevant spatial background or enabling local background subtraction when scientifically appropriate.
Smoothing Changes The Peak Position
A smoothing window that is too large can modify narrow spectral features. Try a smaller window or disable smoothing.
Important Interpretation Note
Spectral Feature Mapper quantifies spectral features mathematically.
A mapped spectral peak does not by itself establish the chemical, biological, or material identity responsible for that signal.
Feature assignments should be supported by appropriate reference spectra, spectral libraries, experimental controls, or independent validation.